Agent as a Tutor: Professional Technical Tutor & Pedagogical Companion

An evidence-calibrated Technical Tutor & Pedagogical Companion skill for AI Coding Assistants (Google Antigravity, Cursor, Windsurf, Claude Code) and web chat platforms (ChatGPT, Claude, Gemini).

Grounded in Cognitive Science principles (Cognitive Load Theory, Desirable Difficulties, Active Retrieval, and the Protégé Effect), this skill transforms your AI assistant into an adaptive personal tutor that fosters genuine, first-principles understanding while respecting your energy, time, and autonomy.

███████╗██╗  ██╗██╗██╗     ██╗     ███████╗
██╔════╝██║ ██╔╝██║██║     ██║     ██╔════╝
███████╗█████╔╝ ██║██║     ██║     ███████╗
╚════██║██╔═██╗ ██║██║     ██║     ╚════██║
███████║██║  ██╗██║███████╗███████╗███████║
╚══════╝╚═╝  ╚═╝╚═╝╚══════╝╚══════╝╚══════╝

The open agent skills ecosystem • strict-tutor v3.1

Quick Start & Installation

Option 1: AI Coding Assistants via npx (Zero Dependencies)

Install directly into your workspace's .agents/skills/strict-tutor/:

# Install to current project root:
npx agent-as-a-tutor .

# Install into a specific subfolder or monorepo package:
npx agent-as-a-tutor ./packages/backend

# Or install via the open agent skills registry:
npx skills add DamPhuQuy/agent-as-a-tutor

Option 2: AI Web Chat Platforms (ChatGPT, Claude, Gemini, LibreChat)

For web interfaces without local CLI or filesystem access, use the single self-contained prompt bundle:

• Open skills/strict-tutor/web-prompt.md.
• Copy its contents and paste into your Custom Instructions, System Prompt, or the initial turn of a chat session.

Core Philosophy: Learning by Intent, Not Forced Interrogation

Unlike rigid tutors that trap you in endless Socratic quizzes or refuse to explain basic concepts, strict-tutor v3.1 follows a Learner-Intent Hierarchy:

| Intent | When You Say | How Tutor Responds | |---|---|---| | EXPLORE | "What is X?", "Cho mình biết khái niệm X" | Delivers a direct, high-signal explanation and concise worked example. No prerequisite gatekeeping. | | UNDERSTAND | "Explain under the hood", "Giải thích sâu cơ chế" | Provides detailed deterministic traces (memory, runtime, threads, storage) with high technical depth. If you request no questions, it honors that request. | | PRACTICE | "Quiz me", "Luyện tập bài tập này", --practice | Offers calibrated micro-challenges and tiered hints (analogy $\to$ search space narrowing $\to$ worked micro-step). | | ASSESS | "Test my transfer", "Kiểm tra mức hiểu" | Evaluates unassisted reasoning and novel transfer scenarios against explicit criteria. |

Pedagogical Tactics & Compatibility Flags

Flags and simulator modes are adaptive tactics, never inescapable traps:

                                  ┌──────────────────────────┐
                                  │   /tutor <topic>         │
                                  │   (Adaptive Mode)        │
                                  └────────────┬─────────────┘
                                               │
             ┌─────────────────────┬────────────┴─────────────┬─────────────────────┐
             ▼                     ▼                          ▼                     ▼
    ┌─────────────────┐   ┌─────────────────┐        ┌─────────────────┐   ┌─────────────────┐
    │  --anchor       │   │  --practice     │        │  --socratic     │   │  --teach-back   │
    │  Domain Anchor  │   │  Schema Coach   │        │  Socratic Mentor│   │  Novice Protégé │
    │  (Primitives)   │   │  (Scaffolding)  │        │  (Boundaries)   │   │  (Alex / LBT)   │
    └─────────────────┘   └─────────────────┘        └─────────────────┘   └─────────────────┘
                                                                                    │
                                                                                    ▼
                                                                           ┌─────────────────┐
                                                                           │  --cards        │
                                                                           │  Flashcard TSV  │
                                                                           └─────────────────┘

1. Adaptive Mode (Default): Balanced explanations and calibrated challenges matching your pacing (/tutor raft consensus).
2. Domain Anchor (--anchor): Grounding in primary codebase files, core primitives, and system invariants.
3. Schema Coach (--practice): Deliberate practice with scaffolded hints. If you get stuck or exhausted, you can ask for the answer directly and it will be provided with full causal reasoning.
4. Socratic Mentor (--socratic / --strict): Stress-tests mental models with counterfactual probes and architectural boundary trade-offs.
5. Novice Protégé / Teach-Back (--teach-back): Reverses roles using the Protégé Effect. You teach Alex, a curious peer.
   • Instant Persona Exit: Say "stop roleplaying", "dừng đóng vai", or ask for an explanation, and Alex drops persona immediately on the very next turn.

Honest Evidence Tracking

Competency is tracked transparently across 4 distinct evidence levels:

• UNASSESSED: Concept not yet addressed.
• INTRODUCED: Explained or demonstrated by the tutor; no independent solution yet.
• ASSISTED: Solved with hints, scaffolding, or after seeing solution elements.
• INDEPENDENT: Solved autonomously in an unassisted, novel problem context.

Epistemic Boundary: Receiving hints or viewing the solution always records the attempt as ASSISTED. Independent mastery is never certified on a problem that was solved with assistance.

Flashcard Export (--cards)

At session wrap-up or whenever requested, the tutor compiles active retrieval flashcards in clean TSV format ready for import into Anki:

# Card ID	Card Type	Front (Prompt)	Back (Target Explanation / Trace)	Tags
CARD-01	Concept	What core problem does Raft consensus solve, and what is its election invariant?	Split-brain prevention; at most one leader can be elected in any given term.	raft,consensus
CARD-02	Boundary	When is Raft linearizable lease read invalid?	During clock drift or an undetected network partition where a split-brain leader continues serving reads without quorum re-check.	raft,boundaries

Note: The tutor generates import-ready card content. Automated spaced repetition scheduling requires dedicated software such as Anki.

Verification & Contract Tests

This repository maintains zero-dependency test coverage using Node.js built-ins:

# Verify web prompt bundle is synchronized and up-to-date:
node scripts/build-tutor-web.cjs --check

# Rebuild web prompt bundle if canonical source files were edited:
node scripts/build-tutor-web.cjs

# Run contract tests (mirror parity, seeds, bundle integrity, CLI smoke):
node --test tests/tutor-contract.test.cjs

Repository Structure

skills/strict-tutor/
├── SKILL.md                          # Canonical skill manifest & intent routing
├── web-prompt.md                     # Generated self-contained web chat prompt
└── references/
    ├── pedagogy-rubric.md            # Evidence hierarchy, 5 dimensions, teach-back rubric
    ├── anti-cargo-cult-rules.md      # Scaffolding ladder, recovery heuristics, exit rules
    └── session-template.md           # Session ledger schema & flashcard TSV template
.agents/skills/strict-tutor/          # Identical mirror for agent environments
scripts/
└── build-tutor-web.cjs               # Deterministic bundle builder (--check support)
tests/
├── tutor-contract.test.cjs           # Deterministic contract & smoke test suite
└── evals/
    ├── tutor-cases.json              # 38 evaluation cases (19 base E/LX + 19 variants)
    └── README.md                     # Replay protocol, severity criteria, and limitations
bin/
└── cli.js                            # Standalone zero-dependency installer

Pedagogical Foundations & Citations

• Dunlosky et al. (2013): Improving Students' Learning With Effective Learning Techniques (Highlighting active retrieval practice and distributed learning).
• Bjork & Bjork (2011): Making Things Hard on Yourself, But in a Good Way: Creating Desirable Difficulties to Enhance Learning.
• Sweller (1988): Cognitive Load Theory (Managing intrinsic and extraneous cognitive load to optimize germane learning).
• Bargh & Schul (1980): The Protégé Effect / Learning by Teaching (Enhanced cognitive consolidation when explaining concepts to an inquisitive peer).
• Risko & Gilbert (2016): Cognitive Offloading (Balancing computational assistance with active internal struggle).

License

MIT © Phu Quy Dam